Earnings Guidance and Managerial Myopia

Mei ChengEller School of Business and Public Administration

University of ArizonaTucson, AZ 85721-0108

Email: meicheng@email.arizona.edu

K.R. SubramanyamMarshall School of Business

University of Southern CaliforniaLos Angeles, CA 90089-1421Email: krs@marshall.usc.edu

Yuan ZhangGraduate School of Business

Columbia UniversityNew York, NY 10027

Email: yz2113@columbia.edu

October 2007

We thank Steven Baginski, Phil Berger, Vicky Dickinson, Jenny Tucker, Paul Zarowin andparticipants at the 2005 New York University Accounting Summer Camp, 2006 FARSconference, and workshops at London Business School, University of Florida and University ofTexas-Dallas for helpful comments and suggestions. We thank Brian Bushee for providingtransient institutional-holdings data, First Call for providing earnings, analyst forecasts, andcompany-issued guidelines data, and I/B/E/S for providing analyst long-term growth forecastsdata. Yuan Zhang thanks Columbia Business School for financial support.

1

EARNINGS GUIDANCE AND MANAGERIAL MYOPIA

ABSTRACT

We examine whether firms that frequently issue quarterly earnings guidance behave myopically,

where myopic behavior is defined as sacrificing long-term growth for the purpose of meeting

short-term goals (Porter [1992]). We find that dedicated guiders invest significantly less in

research and development (R&D) than occasional guiders. This result is robust to controlling for

other determinants of R&D investment as well as to the endogeneity between firms’ guidance

frequency and R&D intensity. We also find that, in comparison to occasional guiders, dedicated

guiders meet or beat analyst consensus earnings forecasts more frequently and they both manage

expectations downward and cut R&D expenditures to achieve this goal. However, we find that

dedicated guiders’ long-term earnings growth rates are significantly lower than those of

occasional guiders. Overall, our results are consistent with dedicated guiders engaging in myopic

R&D investment behavior and meeting short-term earnings targets with possible adverse effects

for long-term earnings growth.

1

EARNINGS GUIDANCE AND MANAGERIAL MYOPIA

Anne Mulcahy, Chairman and CEO of Xerox, calls pressure from Wall Street for short-term

performance "a huge problem" that may be hurting public companies in the long run. "It's

one of the most dysfunctional things going on in the marketplace today," she said during a

Wharton leadership lecture presentation. "I applaud companies that have pulled back from

setting earnings expectations and are trying to reshape the rules of the road."1

1. Introduction

Recently, a number of firms have discontinued the practice of providing quarterly

earnings guidance to the capital markets. For example, Coca-Cola, a company known for

meeting earnings expectations with enviable consistency, announced in 2002 that it would stop

issuing quarterly earnings guidance, but instead provide more information about its progress

towards meeting long-term objectives. Following Coca-Cola, a host of other companies

including Alcoa, AT&T, Clear Channel Communications, Mattel, PepsiCo, and Sun

Microsystems have announced their intention to discontinue providing quarterly earnings

guidance. A recent survey by the National Investor Relations Institute (NIRI) finds that the

percentage of companies providing quarterly earnings guidance has declined from 75% in 2003

to 52% in 2006.2

The benefits of voluntary disclosure to the capital markets—including improved

liquidity, reduced information asymmetry, lower cost of capital, and lower stock volatility—are

widely documented (see Healy and Palepu [2001]). However, firms discontinuing quarterly

earnings guidance are concerned about the potential costs of such disclosures. They argue that

frequent earnings guidance encourages investors and analysts to emphasize meeting short-term

1 See http://knowledge.wharton.upenn.edu/index.cfm?fa=viewArticle&id=1318&specialId=41.2 See http://www.niri.org.

2

earnings targets which fosters myopic managerial behavior that is detrimental to firms’ long-term

growth and value creation. For example, managers at Coca-Cola believe that discontinuing

quarterly earnings guidance will help the company focus on long-term objectives, such as

expanding its business into new markets, without having to worry about meeting short-term

earnings targets (McKay and Brown [2002]). For similar reasons, CFOs interviewed by Graham

et al. [2005] also deplore the culture of providing quarterly earnings guidance and subsequently

having to meet these targets.

Such viewpoints are also shared by prominent regulators and academics who believe that

CEOs must say “no” to the earnings guidance “game” (e.g., see Levitt [2000], Fuller and Jensen

[2002]). Underscoring the urgent importance of this issue an independent commission

established by the U.S. Chamber of Commerce in March 2007 proposed the discontinuance of

quarterly earnings guidance as one of its major recommendations.3 Specifically the commission

recommended that “all public companies should eliminate the practice of providing quarterly

earnings guidance and that companies should instead provide shareholders and Wall Street with

meaningful additional information on their long-term business strategies” (page 77) after

concluding that “[t]he success of American companies in the increasingly competitive global

marketplace depends largely on how well corporate managers execute their long-term business

strategies. Reducing the pressures to meet precise quarterly earnings targets announced by these

very managers is an important first step in shifting the focus of the U.S. capital markets away

from quarterly results and toward the long-term performance of U.S. companies” (page 79).

While there is some anecdotal evidence and many allegations that quarterly earnings

guidance fosters a myopic managerial culture there is no systematic empirical evidence to-date

3 “Report and Recommendations” by Commission on the Regulation of U.S. Capital Markets in the 21st Century,March 2007, US Chamber of Commerce.

3

linking earnings guidance with managerial myopia.4 Therefore, in this paper we examine

whether firms that frequently issue quarterly earnings guidance behave myopically, where

myopic behavior is defined as sacrificing long-term growth for the purpose of meeting short-

term goals through underinvestment in long-term value creating activities (Porter [1992]).

We classify firms as “dedicated guiders” or “occasional guiders” based on their quarterly

earnings guidance frequency during 2001-2003 and examine whether dedicated guiders engage

in more myopic behavior than occasional guiders. Our proxy for myopic behavior is

underinvestment in research and development (R&D). We find that, ceteris paribus, dedicated

guiders invest significantly less in R&D than occasional guiders, which suggests that earnings

guidance is indeed associated with myopic behavior with respect to R&D spending. We

recognize that providing earnings guidance in itself is a managerial choice that can arise

endogenously based on the managers’ propensity to emphasize short-term performance.

Accordingly, we jointly model firms’ decisions to invest in R&D and provide earnings guidance

through a two-stage simultaneous equations’ estimation procedure. We continue to find that

earnings guidance frequency reduces R&D spending after controlling for endogeneity in the

relation between frequent earnings guidance and R&D expenditure.

We next examine the effects of such myopic behavior on the trade-off between short-

term and long-term earnings performance. We find that dedicated guiders are significantly more

likely to meet or beat consensus analysts’ forecasts of current quarter earnings than occasional

guiders both because they manage expectations downward and because they cut R&D

expenditures to avoid missing forecasts. This suggests that dedicated guiders are more actively

involved in playing the “earnings game” and place greater emphasis on achieving short-term

4 The only research examining the relation between disclosure frequency and managerial myopia is Bhojraj andLibby [2005]. However, Bhojraj and Libby examine the effect of mandatory disclosure frequency (i.e., quarterlyversus annual) on managerial myopia in an experimental setting.

4

earnings targets. Further, we find that ceteris paribus, dedicated guiders have significantly lower

earnings growth over the subsequent 2003-2005 period, which is consistent with negative

consequences of myopic behavior on long-term growth and value creation.

Our paper contributes to the literature on several dimensions. First, we add to the

voluntary disclosure literature by investigating possible reasons why companies have recently

discontinued providing quarterly earnings guidance to the capital markets. Chen et al. [2005] and

Houston et al. [2006] investigate causes and consequences of the decisions by firms to

discontinue quarterly earnings guidance. They find that poor short-term performance is an

important motivation for discontinuing earnings guidance. Recognizing potential self-selection

bias associated with this group of firms, we choose a different approach by examining the

relation between earnings guidance practice and myopic decision making for a wide cross

section of firms. We find systematic empirical evidence consistent with anecdotal reports and

claims by managers of companies discontinuing quarterly earnings guidance that such quarterly

earnings guidance forces short-term orientation and impedes investments in long-term value

creation. Our paper thus provides systematic empirical evidence that offers a rationale for this

recent economic phenomenon. In this manner, our paper makes a timely contribution to an

important and current policy issue facing corporate America.

More importantly, our evidence suggests a potential cost of voluntary disclosure:

managerial myopia. While a large literature has documented myriad benefits of voluntary

disclosure to the capital market, such as reduced information asymmetry and improved liquidity,

there is little empirical research examining costs of voluntary disclosure. Our paper is one of the

first to document empirical evidence regarding potential costs associated with frequent quarterly

earnings guidance. Our results have implications for managers who make voluntary disclosure

5

policies as well as for investors who interpret firms’ disclosure information. We note, however,

that our results highlight costs associated with one particular type of voluntary disclosure:

quarterly earnings guidance. We are not implying that managerial myopia is a necessary

consequence of all forms of voluntary disclosure. On the contrary, even firms that have

discontinued explicit short-term earnings guidance have substituted such guidance with

disclosures regarding long-term plans and progress towards long-term goals (e.g., Coca-Cola).

Our paper also adds to the literature on managerial myopia. In a recent survey by

Graham, et al. [2005], managers describe the trade-off between the short-term need to “deliver

earnings” and the long-term objective of making value-maximizing investment decisions.

Consistent with their claim, there is a growing body of empirical evidence showing that

managers are willing to sacrifice long-term value creation to achieve short-term earnings targets.

Prior research documents that firms reduce R&D expenditure in response to concerns regarding

reported income (Baber et al. [1991]) and to cater to the short-term needs of transient

institutional investors (Bushee [1998]). More recently, Bhojraj et al. [2005] document that using

accruals or discretionary expenditures (such as R&D expenditure) to meet or beat analyst

forecasts results in short-term positive impact on firm performance, but long-term

underperformance relative to firms that do not manage earnings to meet forecasts. Along this line

of research, our paper adds a new dimension to the literature on managerial myopia—the

association between quarterly earnings guidance and managerial myopia. To the extent that

quarterly earnings guidance causes managerial myopia, firms have the option to “liberate”

themselves from the earnings guidance game and focus on long-term value creation, as many

have chosen to do in the recent past.

6

2. Motivation and Hypotheses

2.1 MOTIVATION AND THEORETICAL DISCUSSION

2.1.1 Managerial Myopia

U.S. managers have often been criticized for excessive obsession with short-term

performance and resulting myopic investment behavior (e.g., Jacobs [1991], Porter [1992]). Such

short-term focus has been attributed to the willingness of U.S. managers to yield to capital

market pressures by actively participating in the “earnings game” (e.g., Fuller and Jensen

[2002]). For example, Rappaport [2005] observes that investors and managers have a “mutually

reinforcing obsession with short-term performance,” with earnings being the metric of short-term

performance.

Two conditions are required for myopic managerial behavior (Bushee [1998], Stein

[1988, 1989]). First, managers must place sufficient weight on the current market value of their

firm. When managers overweight current market value vis-à-vis future market value, they are

unwilling to wait until the temporary mispricing (if any) based on short-term performance is

corrected and hence unwilling to adopt a long-term perspective. Managers can be motivated to

overweight current stock price because of poorly designed equity incentives (Hall and Murphy

[2003], Rappaport [2005]), takeover threats (Stein [1988]), need to raise capital (Bhojraj and

Libby [2005], Teoh et al. [1998]) or employment concerns (Fundenberg and Tirole [1995],

Graham et al. [2005]). Second, the capital markets should misprice current earnings without

regard to its underlying economics or managers should believe that they do. Markets can

incorrectly price current earnings if investors have a short-term focus (Ellis [2004]) or

misinterpret the persistence of earnings components (e.g., Sloan [1996]). Also, managerial

myopia can result even when the capital markets are efficient as long as managers believe the

7

markets can be fooled (Stein [1989]), which they do (e.g., Graham et al. [2005]). Alternatively,

myopic managerial behavior can also arise when managerial compensation function directly

emphasizes short-term earnings performance, such as meeting earnings targets.

Prior research documents evidence of managerial myopia, primarily with respect to R&D

spending. Some studies show underinvestment in R&D in situations where managers’ horizons

are excessively short-term—for example, Dechow and Sloan [1991] study firms prior to CEO

retirement and Baber et al. [1991] examine loss making firms. Bushee [1998] directly studies the

effects of capital market pressures on myopia, by documenting that myopic R&D spending is

more pronounced in firms that are held by more transient institutional investors. Roychowdhury

[2006] also reports evidence of earnings management through real activities to meet short-term

earnings targets. Finally, Bhojraj et al. [2005] examine the capital market effects of managerial

myopia: they show that firms that meet earnings targets through real earnings management are

able to boost stock price in the short run but suffer adverse price reversals a few years later.

2.1.2 The Role of Earnings Guidance

In this paper we examine the role earnings guidance plays in perpetrating the mutually

self-enforcing obsession of both managers and the capital markets with short-term earnings

performance. As Michael Jensen contends, earnings guidance leads to a short-term mentality for

both companies and investors, thereby shifting the focus from the firm’s fundamentals to bottom-

line earnings (Prince [2005]). Once expectations are managed through earnings guidance,

companies are under intense pressure to meet their earnings forecasts for two reasons. First,

earnings guidance induces investors to place too much emphasis on meeting or beating earnings

targets, resulting in extreme price drops when firms fail to meet targets (Skinner and Sloan

[2002]). Second, managers may face loss of reputation and credibility if they are unable to

8

deliver on their forecasts (Graham et al. [2005]). The pressure to meet short-term earnings targets

in turn precipitates actions that destroy long-term shareholder value creation, such as reduction in

R&D spending or cancellation of marketing campaigns, when otherwise the firm is in danger of

not being able to meet its forecast. Fuller and Jensen [2002] argue that this “earnings game” even

disrupts budgeting and planning processes in organizations. The deleterious effects of short-term

performance obsession arising from the “earnings game” are often cited as one reason why many

companies have recently discontinued the practice of providing quarterly earnings guidance

(Prince [2005]).

While there have been much speculation and discussion based on anecdotal evidence, to

date there is surprisingly no systematic evidence that suggests frequent earnings guidance causes

managerial myopia. Bhojraj and Libby [2005] conduct an experiment where they examine

whether short-term capital market pressures can force managers to make myopic investment

decisions. They also show that the frequency of mandatory reporting of financial performance

(i.e., annual, bi-annual or quarterly) can affect the manner in which managers react to capital

market pressures. However, Bhojraj and Libby do not examine the role of voluntary earnings

guidance in managerial myopia.

2.1.3 The Endogenous Nature of Earnings Guidance

It must be recognized that providing earnings guidance is itself a managerial decision

choice that arises endogenously. Figure 1 schematically illustrates this web of endogenous

relations involving earnings guidance. The decision to issue quarterly earnings guidance can be

viewed as an optimal managerial response that trades off its respective costs and benefits. It is

evident that the choice to issue earnings guidance arises from managers’ preoccupation with

short-term (earnings) performance. The benefits of focusing on short-term earnings performance

9

could arise from various factors such as poor compensation policies and equity incentives (Hall

and Murphy [2003], Matsugana and Park [2001]), need to raise capital (Bhojraj and Libby

[2005]), job-tenure concerns (Fundenberg and Tirole [1995]) and short managerial horizon

(Dechow and Sloan [1991]). In addition, poor financial performance can increase litigation risk,

which in turn can induce voluntary disclosure (Skinner [1994]). By issuing frequent earnings

guidance, however, managers sensitize the stock market to focus on quarterly earnings’

performance and get entrenched in the “earnings game” of setting, managing and meeting

market’s earnings expectations to which the market responds, which in turn leads to the

“mutually reinforcing obsession with short-term performance” on the part of both the investors

and managers. (This endogenous loop is depicted with bold arrows in Figure 1). An extreme

focus on short-term earnings performance also creates unhealthy pressures on managers to meet

short-term earnings targets, which may lead to myopic underinvestment in long-term value

creating activities, such as R&D.5

Many of the factors that induce managers to issue frequent earnings guidance can also

directly (and independently) cause managerial myopia. For example, some of the factors that

cause managers to emphasize short-term performance (such as poor compensation policies or job

concerns) may directly pressure managers to meet short-term earnings targets and hence indulge

in myopic behavior, independent of the decision to issue earnings guidance. In addition, poor

financial performance can also directly cause underinvestment through financial constraints.

Finally, poor financial performance may increase litigation risk, which may also induce

managers to emphasize short-term performance and hence behave myopically.

5 Markets obsession with short-term performance may also arise through the presence of transient institutionalinvestors, which in turn may lead to managerial focus on short-term performance and resulting managerial myopia(Bushee, 1998).

10

If earnings guidance is an entirely endogenous choice and if the exogenous determinants

of earnings guidance are hypothesized to independently cause managerial myopia, then there is a

concern whether the fundamental relation that we seek to examine (i.e., that frequent earnings

guidance can cause managerial myopia) exists at all. In other words, theoretically is there a

component of dedicated earnings guidance that is orthogonal to its various determinants that

causes myopic investment? We argue that such a component exists and that it arises because of

the mutually reinforcing nature of the relation between managers and markets involved in the

“earnings game”. Specifically, although the initial choice of providing earnings guidance may be

triggered by exogenous factors (such as a poorly designed compensation schemes), regular

earnings guidance entrenches the manager and the markets into the mutually reinforcing

“earnings game” from which it is costly for mangers (corporations) to get out of.

2.2 HYPOTHESES

Myopia refers to sub-optimal underinvestment in long-term projects for the purpose of

meeting short-term goals (e.g., Porter [1992]). This definition has three aspects: (1) there should

be underinvestment in long-term value creation projects, such as R&D; (2) the underinvestment

should occur with the objective of meeting short-term goals, such as meeting/beating analyst

earnings forecasts; and (3) such underinvestment must be sub-optimal in the sense of impairing

long-term growth and value creation. The first aspect reflects myopic behavior, and the second

and third aspects (i.e., short-term and long-term performance) reflect the motivation and

consequence of myopic behavior. Our hypotheses and associated empirical tests address each of

the three aspects in order. We motivate our hypotheses below.

11

For examining whether earnings guidance results in underinvestment in long-term

projects, we examine differences in R&D spending levels between dedicated guiders and

occasional guiders. R&D expenditures are particularly appropriate for examining myopic

behavior because they have important but opposite implications for short-term and long-term

performance. First, since R&D expense is immediately expensed under U.S. GAAP, under-

investing in R&D will boost short-term earnings. Second, adequate investment in R&D is crucial

for the viability and success of firms in many industries. However, gains from R&D are realized

only in the long-term—in contrast to, say, advertising expense that can boost sales in the short-

term. Prior research (e.g., Dechow and Sloan [1991], Bushee [1998]) uses R&D expenditures to

evaluate myopic behavior. Also survey evidence (Graham et al. [2005]) suggests that managers

use R&D expenditures to play the “earnings game”. Accordingly, our first hypothesis (in

alternative form) is:

H1A: Ceteris paribus, dedicated guiders invest less in R&D than occasional guiders.

Recent research provides evidence that there is pressure on firms to avoid missing analyst

earnings forecasts and that such pressure has intensified in recent years (e.g., Matsumoto [2002],

Kasznik and McNichols [2002], Skinner and Sloan [2002]). In particular, Brown and Caylor

[2005] find that in recent years (since mid-1990s), managers seek to avoid negative quarterly

earnings surprises more than to avoid either quarterly losses or quarterly earnings decreases.

Therefore, if dedicated guiders are more prone to play the myopic “earnings game”, we should

observe that such firms are more likely to meet/beat analyst forecasts. Accordingly, our second

hypothesis (in alternative form) is:

H2A: Ceteris paribus, dedicated guiders meet or beat analysts’ quarterly earnings forecasts morefrequently than occasional guiders.

12

Finally, the underinvestment in R&D is myopic only if it is suboptimal, i.e., that the

underinvestment leads to some form of value destruction over the long run. Therefore, if the

lower R&D spending by dedicated guiders is myopic, i.e., sub-optimal, it must be reflected in

lower long-term growth and value creation. Accordingly, in our third hypothesis (in alternative

form), we examine growth in earnings subsequent to our sample period:

H3A: Ceteris paribus, dedicated guiders have lower long-term growth in earnings thanoccasional guiders.

It is important to note that all our hypotheses are valid only under the ceteris paribus

assumption. As we note earlier, earnings guidance arises endogenously. Accordingly, it is

important to control for this endogenity in our empirical analyses. We return to this issue in more

detail in Section 4.

3. Sample and Descriptive Statistics

Regulation Fair Disclosure (FD), which became effective October 2000, changed the

nature of corporate earnings guidance by prohibiting managers from privately communicating

with select market participants (e.g., Heflin et al. [2003]). Accordingly our sample starts from

2001 to ensure that our results are not contaminated by private earnings guidance issued during

the pre-FD period, which cannot be identified through public sources such as the CIG (Company

Issued Guidelines) database from First Call, our source of earnings guidance data. Our guidance

frequency measurement period ends at 2003 because we need the subsequent period to perform

the long-term performance test. Our initial sample includes all firm-quarters in the “Actuals”

database of First Call with fiscal quarters ending between 2001 and 2003. We next require that

the following information be available in various First Call databases: (1) actual earnings per

share and earnings announcement date for the firm-quarter; (2) number of analysts and mean

13

analyst forecasts, both measured at the last summary statistics calculation prior to the

corresponding earnings announcements; and (3) the stock-split adjustment factor. We delete

firm-quarters with earnings announcement dates either before or 45 days after the fiscal quarter

end. We also delete firm-quarters if the last summary statistics calculation occurs after the

earnings announcement. We make these deletions to eliminate data errors or other irregularities.

This selection procedure yields an initial sample of 94,522 firm-quarters.

We next obtain SIC information from Compustat and classify our sample firms into 12

industries according to Fama and French [1997]. Because of their peculiarities, we exclude

utilities (SIC code between 4900 and 4949) and financial services (SIC code between 6000 and

6999). Finally, to ensure that the quarterly earnings guidance practices we observe reflect the

firms’ disclosure policies over a reasonable period of time, we require that our sample firms have

data for all twelve quarters during the 2001-2003 period. This process leads to a sample

representing 1,965 distinct firms in 10 industries. The number of firms varies from 51 in the

chemicals industry to 557 in the business equipment industry, as shown in Table 1.

We obtain quarterly earnings guidance information for our sample from the CIG database

in First Call.6 We define quarterly earnings guidance frequency as the number of quarters (out of

the 12 quarters during 2001-2003) for which the firm makes at least one earnings forecast. We

do not consider guidance issued more than 90 days before the corresponding fiscal quarter-end to

ensure that the guidance represents short-term earnings forecasts for the current quarter. We also

eliminate observations with guidance issued more than 45 days after the fiscal quarter-end to

avoid any potential data errors.

6 To check the comprehensiveness and accuracy of First Call’s coverage of earnings guidance, we randomly select 5dedicated firms and 5 occasional firms in our sample and compare the CIG data in First Call with the PR Newswiredata in Lexis-Nexis. We find that in general, the CIG data covers more earnings guidance than the PR Newswire,and that for the guidance that we are able to find in both data sources, they generally report the same guidance datesand earnings estimates.

14

Table 1 provides descriptive statistics of quarterly earnings guidance frequency for our

sample. Overall, our sample firms’ quarterly earnings disclosure frequency (out of 12) has a

mean of 4.02 and a median of 3. The first quartile is 1 while the third quartile is 7. Out of the

1,965 sample firms, 480 firms (24%) do not provide any quarterly earnings guidance during our

sample period and 72 firms (4%) issue earnings guidance every quarter. This distribution

suggests wide variation in the propensity of firms to provide quarterly earnings guidance.

We also examine variation in quarterly earnings guidance frequency across industries.

There is considerable variation in guidance frequency across different industries. Firms in the

wholesale and retails industries have the highest guidance frequency with mean of 5.93 (median

of 7) quarters—out of 12 quarters—having at least one earnings forecast. In addition, firms in the

industries of chemicals, consumer non-durables and business equipment provide relatively

frequent quarterly earnings guidance as well. On the other hand, firms in the energy sector have

the lowest guidance frequency with mean of 1.98 (median of 0) quarters, followed by

telecommunications and healthcare industries. The within-industry standard deviations range

from 3.00 to 4.12, suggesting that there is also within-industry variation in guidance frequency.

Our empirical analyses require classifying our sample firms into different groups based

on their propensity to issue quarterly earnings guidance. As our descriptive evidence suggests

significant variations across different industries, we perform our classifications by industry.

When a firm’s quarterly earnings guidance frequency is in the top (bottom) third of its respective

industry’s frequency distribution, we classify the firm as a dedicated guider (occasional guider).7

In subsequent empirical analyses, we exclude firms whose guidance frequency is in the middle

third of the corresponding industry’s distribution as their propensity to issue quarterly earnings

7 Occasional guiders also include non-guiders, i.e., those firms that never issue any quarterly earnings guidanceduring our sample period.

15

guidance is ambiguous. Thus, our final sample reduces to 1,406 distinct firms, including 669

dedicated guiders and 737 occasional guiders.

4. Guidance Frequency and R&D Expenditure

4.1 UNIVARIATE RESULTS

We measure R&D expenditure intensity by RDX, which is R&D expenditure for the year

deflated by the total assets at the beginning of the year.8 We use annual data to conduct our

analyses because firms typically report R&D expenditures only on an annual basis. Table 2 Panel

A presents univariate results for R&D expenditure comparisons between dedicated and

occasional guiders. The mean investment in R&D by dedicated guiders is 4.7% of total assets

compared to 8.4% by occasional guiders and the difference is significant at p < 0.01. This result

suggests the R&D intensity of dedicated guiders is just more than half that of their occasional

counterparts. Although the median RDX for dedicated guiders is also statistically lower than that

of occasional guiders, the difference is much smaller.

Because R&D intensity varies significantly by industry, we also report in Panel A means

and medians of industry-adjusted RDX (ADJRDX). Consistent with unadjusted numbers,

dedicated guiders (mean -0.018, median -0.010) invest significantly less in R&D than occasional

guiders (mean 0.016, median -0.004) after we control for average spending in each industry.

Note the mean difference (0.034) is the ballpark of the difference without industry adjustment

but the median difference (0.014) is now economically significant.

8 We replicate our tests using R&D expenditure deflated by lagged sales and find qualitatively similar results.

16

4.2 MODELLING R&D INTENSITY AND GUIDANCE FREQUENCY

The discussion in Section 2 suggests that both R&D intensity and earnings guidance

propensity are jointly influenced by a number of factors. Accordingly, we simultaneously model

R&D intensity and guidance frequency through the following system of equations (firm-year

subscripts omitted):

RDX=0+iINDi+1DEDICATED+2LAGRDX+3LAGDEDI+4BM+5LOGMV+6EP

+7AGE+8ADJROA+9NADJROA+10LOSS+11CAPX+12ADJFCF+13DILUTION

+14EQOFFER+15%TINST +16NUMANA+17VALRE+ (1)

Prob(DEDICATED=1)=f(0+iINDi+1RDX+2LAGDEDI+3MAFE0+4MSTD0

+5LOGMV+6NUMANA+7PIN+8EP+9EXPDAM+10LOSS+11DILUTION

+12EQOFFER+13%TINST+14VALRE+) (2)

where

RDX = R&D expenses (Compustat data#46) for the year scaled by total assets(Compustat data#6) at the beginning of the fiscal year,

IND = industry dummy variables,DEDICATED = one (zero) if the firm’s quarterly earnings guidance frequency is in the top

(bottom) third of their respective industry’s distribution of guidancefrequency based on the observation from year 2001 to 2003,

PREDDEDI = the predicted value of dedication, which is the fitted value (i.e., a continuousmeasure) from the first-stage logistic regression of DEDICATED on allexogenous variables from Model (1) and Model (2) together,

PREDRDX = the predicted value of R&D expenses, which is the fitted value from thefirst-stage ordinary least square regression of RDX on all exogenousvariables from Model (1) and Model (2) together,

LAGRDX = R&D expenses (Compustat data#46) at the end of year 2000 scaled by totalassets (Compustat data#6) at the beginning of year 2000,

LAGDEDI = the rank of the sample firms’ quarterly earnings guidance frequency (intothree groups) within their respective industry’s distribution based on datafrom 1998 to 2000,

BM = ratio of book value of equity (Compustat data#60) to market value of equity(Compustat data#25 x data#199) at the end of the year 2000,

LOGMV = log of market value at the end of the year 2000,EP = earnings per share (Compustat data#58) divided by price per share

(Compustat data#199) at the end of the year,AGE = log of firm age,ADJROA = R&D-adjusted return on assets measured as net income before extraordinary

items (Compustat data#18) plus 0.65 times R&D expenses (Compustat

17

data#46) divided by total assets (Compustat data#6) at the beginning of theyear,

NADJROA = ADJROA if ADJROA is negative and 0 otherwise,LOSS = 1 if net income before extraordinary items (Compustat data#18) is negative

and 0 otherwise,CAPX = capital expenditure (Compustat data#30) for fiscal year scaled by total

assets (Compustat data#6) at the beginning of the fiscal year,ADJFCF = R&D adjusted free cash flow measured using income (Compustat

data#18) minus the change of current assets (Compustat data#4) plus thechange of current liabilities (Compustat data#5) plus the change of cash andshort-term investments (Compustat data#1) minus the change of short-termdebt (Compustat data#34) plus depreciation (Compustat data#14) minuscapital expenditure (Compustat data#30) plus 0.65 times R&D expenses(Compustat data#46) scaled by total assets (Compustat data#6) at thebeginning of the fiscal year,

DILUTION = the difference between basic earnings per share (Compustat data#58) anddiluted earnings per share (Compustat data#57) deflated by the dilutedearnings per share (Compustat data#57) at the end of the year,

EQOFFER = total net value of new equity issuance (Compustat data#108) during 2001 to2003 scaled by total assets at the end of the year 2000,

%TINST = percentage of transient institutional holding at the end of the year fromCDA/Spectrum Institutional (13f) Holdings database according to Bushee(1998),

NUMANA = number of analysts following at the end of the calendar year from First CallSummary database,

VALRE = value relevance proxy, measured as negative one times the squared residualfrom the following regression:

0 1 2 3 4*RET NIBE LOSS NIBE LOSS NIBE .The regression is estimated by three, two, or one-digit SIC code conditionalon having at least 10 firms in each group. RET = market adjusted returnover the fiscal year; NIBE = net income before extraordinary items(Compustat annual data#18) scaled by the beginning market value of equity(Compustat annual data#25*Compustat annual data#199); LOSS =1 ifNIBE is negative, zero otherwise; NIBE = the change in net income beforeextra-ordinary items scaled by beginning market value of equity. Since wemultiply by -1 larger values of VALREL indicates greater value relevance,

MAFE0 = average absolute analyst forecast error measured as the absolute value of thedifference between actual earnings per share and the mean forecast at thebeginning of the quarter scaled by price at the quarter end over the currentyear,

MSTD0 = average standard deviation of analyst forecasts measured as the standarddeviation of the beginning of the quarter forecasts scaled by price at thequarter end over the current year,

PIN = a measure of probability of information-related trading estimated using theEasley, Hvidkjaer and O'hara (2002) trading model for the current year,

18

EXPDAM = expected loss damage following Skinner (1997) to measure potentiallitigation costs calculated as –MVE*MKTADJRET*(1-(1-VOL)**N));MVE=market value at the beginning of the year; MKTADJRET=marketadjusted return over the year; VOL=average percentage of the firm’s sharestraded over the year; N=day period between the end of the year and thebeginning of the year.

4.2.1 Modeling R&D Intensity

Our R&D model (Model 1) is largely based on that in Bushee [1998] and Berger [1993]

and augmented by additional variables that control for various incentives for myopic managerial

behavior. The dependent variable, RDX, is annual R&D expenditure deflated by beginning total

assets. Our variable of interest is DEDICATED, which is an indicator variable capturing whether

a firm is a dedicated or an occasional guider. If dedicated guiders behave myopically, we expect

the coefficient on DEDICATED in Model (1) to be significantly negative. In other words, we

expect dedicated guiders to invest significantly less in R&D, ceteris paribus.

We include industry dummies INDi because of the considerable variation in R&D

intensity across different industries. Also, similar to Berger [1993], we include lagged R&D

(LAGRDX) as an additional independent variable. The inclusion of lagged R&D controls for

cross-sectional variation in R&D intensity and makes our model similar to a “changes”

specification without constraining the coefficient. We also include lagged guidance frequency

(LAGDEDI) to control for a firm’s general propensity to provide earnings guidance. Specifically,

LAGDEDI is constructed similarly as DEDICATED except that it is measured over the prior

three-year period, i.e., over 1998-2000.

We categorize the remaining independent variables in Model (1) into four categories:

growth, financial constraints, managerial short-term incentives and market short-term incentives.

First, we include variables to reflect firms’ growth opportunities both because firms with growth

opportunities are more likely to invest in R&D and because the costs of under-investing in R&D

19

are expected to be higher for such firms (Myers [1984], Berger [1993]). Specifically, we expect

RDX to be negatively related to book-to-market ratio (BM), firm size (LOGMV), earnings-to-

price ratio (EP) and firm age (AGE).

Second, we include variables that capture financial constraints on R&D spending. R&D

is essentially discretionary investment and therefore we expect that financial constraints will

adversely affect the level of R&D expenditure.9 At the outset, we expect the level of R&D

expenditure to depend on the firm’s financial performance as firms with poor financial

performance are likely to be more financially constrained. Accordingly, we include return on

assets adjusted for after-tax R&D expense as an additional control variable (ADJROA). We also

allow for different coefficients on negative and positive return on assets by including NADJROA,

which is ADJROA if it is negative and zero otherwise. We also include a loss indicator that takes

value 1 if the firms reports negative earnings (LOSS), as an additional measure of financial

performance. Additionally, we expect R&D expenditure to be negatively related to capital

expenditure (CAPX) because firms with higher capital expenditure are likely to have less funds

available for R&D (Bushee [1998]). We also include a measure of free cash flow adjusted for

R&D effects (ADJFCF) because it proxies for a company’s financial constraints (or lack

thereof).

The third set of control variables attempt to capture managerial incentives for focusing on

short-term performance. Prior studies suggest that managers endowed with large amounts of

stock options place greater emphasis on short-term performance (e.g., Cheng and Warfield

[2005]). To measure the extent of in-the-money employee stock option ownership, we construct

a measure that is defined as the difference between basic and diluted earnings per share deflated

9 For example, there is burgeoning literature in corporate finance that argues that financial constraints can adverselyaffect firms’ propensity for capital investment (e.g., Fazzari, Hubbard and Petersen, 1988; Kaplan and Zingales,1997).

20

by diluted earnings per share (DILUTION). Because employee stock options are often the most

significant dilutive security, our measure captures the extent of in-the-money employee stock

option ownership. We also include a measure of another managerial short-term incentive that

arises from equity offering (EQOFFER) because firms have incentives to focus on short-term

performance in order to decrease cost of capital.

Finally, we include proxies for various market factors that may encourage or constrain

managerial myopia. Bushee [1998] shows that transient institutional investors emphasize short-

term performance and encourage myopic reductions in R&D spending. Accordingly, we expect

the level of R&D investment to be negatively related to the percentage ownership by transient

institutional investors (%TINST). We also predict that financial analysts play a monitoring role,

which might mitigate managerial myopia (Jensen and Meckling [1976]). Hence we include the

number of analysts following the firm (NUMANA) in the model. Finally, we include a measure

for the value relevance of earnings (VALRE). We predict that, because capital markets place

greater weight on more value-relevant earnings, managers of such firms might myopically

emphasize short-term earnings performance.

4.2.2 Modeling Disclosure Frequency

In Model (2), we model guidance frequency based on the fairly extensive literature on

voluntary disclosure incentives (e.g., Lang and Lundholm [1993], Skinner [1994], Healy and

Palepu [2001]). Our dependent variable is DEDICATED which is a dummy variable that takes on

the value 1 for dedicated guiders and 0 otherwise. We include RDX as an independent variable

when modeling guidance frequency because R&D investments and frequency of earnings

guidance may be reciprocally caused. We include our lagged guidance frequency measure

21

(LAGDEDI) to control for firms’ historical propensity to provide frequent guidance. Finally, we

include industry indicator variables (IND) to control for industry-specific effects.

We categorize our remaining control variables into the following groups: expectations

adjustment, demand for information, litigation risk, managerial short-term incentives and market

short-term incentives. First, we follow Ajinkya and Gift’s [1984] argument that managers

provide earnings guidance in order to align the market expectations with their own. We include

the mean quarterly absolute analysts’ earnings forecast error (MAFE0) and standard deviation of

analysts’ earnings forecasts (MSTD0) as proxies for the demand for expectations adjustment.

Second, we include several variables that capture demand for information. We include the log of

market value of equity (LOGMV) and number of analysts covering the firm (NUMANA) because

we expect managers of larger firms and firms with higher analyst coverage to face greater

demand/pressure to provide information to the market. At the same time, we also expect that

investors of firms with higher information asymmetry will demand more guidance from

managers. We measure information asymmetry by probability of informed trading (PIN) and the

earning-price ratio (EP).

Third, following Skinner [1997], we include a proxy for litigation risk that attempts to

measure potential damages from litigation as the dollar amount of investors’ losses measured

over the prior year (EXPDAM).10 While litigation risk is expected to affect corporate guidance

policy, we do not make a directional prediction for the effect of EXPDAM because as Healy and

Palepu [2001] suggest, litigation risk can have two offsetting effects on managers’ disclosure

decisions. First, legal actions against managers for inadequate or untimely disclosures can

encourage firms to increase voluntary disclosure to avoid surprising the market with bad news

10 The measure in Skinner [1997] requires a class action period as his sample firms all are subject to actual classaction lawsuits. Since our sample is a more general sample without actual litigation, we use a one-year period priorto the year as the measurement period of the expected damage variable.

22

(Skinner 1994). Second, litigation can potentially reduce managers’ incentives to provide

disclosure, particularly of forward-looking information such as earnings guidance that can be

subsequently verified. We also include the loss indicator (LOSS) to control for potential litigation

risk, although we again do not predict the direction of its effect.11

Our fourth and fifth sets of control variables, managerial short-term incentives and

market short-term incentives, are similar to those included in the R&D model. The rationale for

including these variables is that short-term incentives (internal or market related) will cause the

manager to focus more on short-term earnings, thereby inducing them to engage in expectations

management and therefore provide frequent quarterly earnings guidance. Therefore we include

DILUTION and EQOFFER (managerial short-term incentives) and %TINST and VALRE (market

short-term incentives) in the model.

4.3 MULTIVARIATE REGRESSION RESULTS

4.3.1 Single-Stage Multiple Regression

We first estimate Model (1) using single-stage OLS regression and report the results in

Panel B of Table 2. We remove all observations with absolute studentized t-values greater than 2

to control for outlier effects. As hypothesized, the coefficient on DEDICATED is significantly

negative at -0.004 (p=0.02), suggesting that dedicated guiders, after controls, on average under-

spend their occasional counterparts on R&D by 0.4% of total assets.12 This result is consistent

with dedicated guiders under-investing in R&D, although the magnitude of the difference is

order-of-magnitude lower than that in the univariate analysis.

11 LOSS also proxies for performance. Prior studies find a positive association between performance and disclosure(Lev & Penman [1990], Lang and Lundholm [1993]).12 The reported p-values are based on simple t-statistics from OLS regressions. We also estimate Newey-Westcorrections to all our regressions and find similar results on all our key variables.

23

Regarding the control variables, as predicted, firm-years with higher lagged R&D

investment (LAGRDX), lower book-to-market ratio (BM), younger age (AGE), higher adjusted

return on assets (ADJROA), lower capital expenditure (CAPX), lower dilution factor

(DILUTION), less transitory institutional holding (%TINST), lower value relevance of earnings

(VALRE) tend to invest more on R&D. Inconsistent with expectations, firm-years with higher

earnings price ratio (EP), negative earnings (LOSS), more net equity issuance (EQOFFER) tend

to invest more in R&D. The coefficient on NADJROA is significantly negative. All other control

variables are insignificant.

4.3.2 Two-Stage Regression Controlling for Endogeneity

The OLS regression is misspecified if R&D expenditure and guidance frequency are

endogenously determined. Therefore, based on Models (1) and (2), we conduct the “omitted

variables” version of the Hausman [1978] test of endogeneity between RDX and DEDICATED

(see Beatty et al. [1995]). Unreported results suggest that RDX and DEDICATED are indeed

endogenously determined. Accordingly we simultaneously estimate RDX and DEDICATED

using the system specified in Models (1) and (2).13

To estimate the simultaneous equations, we employ a two-stage procedure recommended

by Nelson and Olsen [1978] (see DeFond et al. [2002] for an application in an accounting

context). Specifically, in the first stage, we model RDX and DEDICATED separately using all

exogenous variables in Models (1) and (2). We use OLS estimation for Model (1) and logit

estimation for Model (2) and generate predicted values for RDX and DEDICATED. In the second

13 The reported results are based on the full sample with data availability. We also control for outliers in all our two-stage analyses in the following manner. In the RDX model (OLS regression) we exclude observations with absolutestudentized t-values greater than 2 and in the DEDICATED model (logit regression) we exclude all observationswith Pearson-stat greater than 2. The inferences are unchanged.

24

stage, we replace the independent variables RDX and DEDICATED with their respective

predicted values in Models (2) and (1) respectively (i.e., PREDRDX and PREDDEDI).14

Larcker and Rusticus [2005] highlight two potential problems with the two-stage

regression procedure. The first problem is that of weak instruments; that is, the instrumental

variables (IV) are weakly correlated with the endogenous variable. Such a problem can arise

when the exogenous variables selected by the researcher in the first stage are unable to

adequately explain the variation the endogenous variable. Weak instruments can cause the

coefficient on the endogenous variable to still be biased in the direction of the single-stage OLS

regression. We perform the following tests to examine the strength of our IVs for DEDICATED.

First, we examine the goodness-of-fit statistics for the logit estimation of Model (2) (i.e., where

DEDICATED is the dependent variable). Both the Likelihood Ratio Test and the Wald Test

reject the null hypothesis of lack-of-fit at p < 0.01. Second, we estimate the correlation between

the endogenous variable (i.e., DEDICATED) and the predicted value of DEDICATED from the

first stage logit regression. The Pearson correlation is 0.63 and significant at p < 0.01. Together,

these results suggest that it is unlikely that we have a weak instrument problem in our two-stage

regression.

A second potential problem with the two-stage method, according to Larcker and

Rusticus [2005], is that of semi-endogenous instruments, i.e., the chosen instruments are not

entirely exogenous. If the instruments themselves are endogenous, it leads to bias in the two-

stage coefficients. In case of over-identified models such as ours (i.e., where the number of

instruments exceeds the number of endogenous regressors), it is possible to regress the estimated

14 Amemiya [1978] shows that although the coefficients from the Nelson-Olsen second-stage equation areasymptotically unbiased, they are not the most efficient estimators and the covariance matrix of the estimators arebiased. He proposes a method for deriving more efficient estimators and a correction to the covariance matrix that isunbiased. Similar to DeFond et al. [2002], we do not implement this correction because the extent of this bias isexpected to be small and is unlikely to affect our inferences.

25

error term in the second-stage equation on the instruments to determine whether the chosen

instruments are valid, i.e., exogenous (see Hausman [1978]). In particular, the R-squared from

this regression can be used to perform a Chi-squared test; a significant Chi-squared statistic

indicates that the instruments are potentially endogenous. Accordingly, we regress the residuals

from the second stage R&D model on our exogenous variables (instruments). The coefficients on

all but one of our instruments are not significantly different from zero.15 Also the Chi-squared

test is insignificant at conventional levels (p > 0.99).16 Thus, our tests indicate that our

instruments are reasonably exogenous.

Overall, our tests detect the presence of endogeneity between RDX and DEDICATED, but

suggest that our set of instruments are neither semi-endogenous nor weak in explaining the

endogenous variables. Accordingly, we implement the two-stage regression procedure. Results

are reported in Panel C of Table 2. Consistent with our hypothesis, the coefficient on

PREDDEDI (i.e., the predicted value for DEDICATED from the first-stage estimation) remains

significantly negative at -0.048 (p < 0.01). This suggests that after controlling for potential

endogeneity, dedicated guiders on average spend 4.8% of total assets less on R&D than

occasionally guiders; the magnitude of the effect increases by an order-of-magnitude relative to

the simple OLS regression reported in Panel B of Table 2, and becomes substantially similar to

that in univariate analysis. This result suggests that the simple OLS coefficient was biased

downward because of endogeneity.

15 The only instrument that is significant in this regression is PIN. Accordingly, in sensitivity analysis, we re-estimate the two-stage regression after excluding PIN from the system of equations. Our results are qualitativelysimilar.16 Specifically, nR2 ~ χ2 (M – N) where n = sample size, R2 = the R-squared from the above regression and M and Nare number of instruments and endogenous variables respectively. In our case, n = 2655; R2 = 0.17%; M = 29 and N= 2. This gives us a Chi-squared value of 4.51 with 27 degrees of freedom, which is equivalent to a p-value of 0.99.

26

In addition, we do not find a significant effect of RDX on DEDICATED, suggesting that

R&D expenditure levels do not influence guidance frequency directly. Regarding the control

variables, we find that the results for model (1) are largely consistent with those in simple OLS

regression with the following exceptions. The coefficients on lagged value of dedication

(LAGDEDI) and analyst coverage (NUMANA) now become significantly positive. Equity

dilution factor (DILUTION) and transitory institutional holding are insignificant.17 Finally, we

note that the adjusted (pseudo) R-squared for the two equations are 68.76% and 49.73%

respectively, which suggests that we have modeled the joint decisions for R&D expenditures and

guidance frequency reasonably well.

In summary, our results are consistent with our hypothesis (H1A) that ceteris paribus

firms that provide frequent quarterly earnings guidance are likely to invest less in R&D than

those who provide guidance occasionally. Our results are robust to controlling for other factors

that affect R&D spending and to controlling for the endogeneity between R&D expenditure and

guidance frequency.

5. Guidance Frequency and Short-Term Performance

5.1 UNIVARIATE ANALYSIS

To test H2A, we first compare the relative frequency with which dedicated and occasional

guiders meet or beat analysts’ quarterly earnings forecasts. We study the ability to meet or beat

analyst expectations at two points in time: at the beginning of the quarter and prior to the

earnings announcement. Accordingly, we define two indicator variables that switch on when a

17 We note that the insignificant coefficient on %TINST is not necessarily inconsistent with Bushee [1998] who findsfirms with higher transient institutional investors are more likely to cut R&D. Bushee examines the probability ofcutting R&D while we in effect examine both the magnitude and the direction of changes in R&D by modeling thelevel of R&D investment controlling for its lagged level. In addition, we have other control variables that are notpresent in Bushee’s model and we estimate a simultaneous equation system, which may also affect the results. The%TINST variable has a significantly negative coefficient in our simple OLS regression reported in Panel B of Table2, consistent with Bushee [1998].

27

firm meets/beats expectations based on the first consensus forecast after the beginning of the

quarter (MBE0) and the last consensus forecast before the actual earnings announcements

respectively (MBE). Panel A of Table 3 shows that only 53% of dedicated guiders would meet or

beat analyst consensus forecasts measured at the beginning of the quarter (MBE0=1 for 53% of

the dedicated guiders), in comparison to 51% of occasional guiders. The difference, although

statistically significant, is small in magnitude. In contrast, by the time of earnings

announcements, dedicated guiders meet or beat market expectations 85% of the time (MBE=1 for

85% of dedicated guiders), compared to only 71% for the occasional guiders. This difference is

statistically significant at better than 0.01 level and the magnitude is considerably larger. Thus,

the evidence supports our hypothesis that dedicated guiders are more likely to meet or beat

quarterly analyst forecast at the time of earnings announcements.

To shed further light on the role of earnings guidance in meeting short-term earnings

targets (through managing the market’s earnings expectations downward18), in Panel B of Table

3 we report the interaction between MBE0 and MBE for dedicated guiders and occasional guiders

separately. While the distribution of MBE is similar between the two groups of firms conditional

on MBE0=1, a much higher percentage of MBE0=0 firms eventually meet or beat analyst

consensus at the time of earnings announcement (i.e., MBE=1) for dedicated firms (70%) than

for occasional firms (47%). A Chi-squared test suggests that this difference is statistically

significant (p < 0.01). Thus, while the results in Panel A suggest that dedicated guiders have

greater emphasis on meeting or beating short-term analyst forecasts, the evidence in Panel B

18 Expectations management does not preclude the possibility of the firm using other approaches, such as earningsmanagement or real operations management (i.e., cutting R&D expenditures or foregoing positive-NPV projects) tomeet or beat market expectations. For example, Kasznik [1999] finds that managers who issue earnings forecastsalso manage reported earnings toward their forecasts; Matsumoto [2002] finds that managers use both earningsmanagement and expectations management to avoid negative earnings surprises. In Section 5.3 we provide a directtest suggesting that dedicated firms use R&D expenditure to meet/beat expectations.

28

suggests this is mainly driven by dedicated firms that would have missed analyst forecasts based

on consensus forecasts early in the quarter. Overall our results suggest that dedicated guiders

participate more actively in the earnings meeting/beating “game”, in particular through managing

earnings expectations.

5.2 REGRESSION ANALYSIS

We next estimate a logistic regression that controls for different factors affecting the

probability of meeting or beating quarterly analyst consensus earnings forecasts. We follow

Matsumoto [2002] who models this probability based on firms’ incentives to avoid negative

earnings surprises. Specifically, we estimate the following logistic regression model (firm-

quarter subscripts omitted):

Prob(MBE=1)=F(0+iINDi+1DEDICATED+2ML%INST+3MLRDX

+4MLLABR+5QLOSS+6GROWEST+7POSUE+8QLOGMV+9AFE0+)

(3)

MBE and DEDICATED are as defined earlier. DEDICATED is our variable of interest. To the

extent that firms frequently issuing earnings guidance have more emphasis on short-term

objectives such as meeting or beating market expectations, we expect the coefficient on

DEDICATED to be significantly positive. Matsumoto [2002] shows that firms with higher

institutional ownership (ML%INST), higher R&D expenditures (MLRDX), and higher labor

intensity (MLLABR) are more likely to meet or beat analyst consensus. Accordingly, we control

for these variables. As in Matsumoto [2002], ML%INST is the average percentage of institutional

ownership over the previous four quarters where institutional ownership information is obtained

from CDA/Spectrum Institutional Holdings (13f) database; MLRDX is the average R&D

29

expenditure deflated by total assets over the previous four quarters19 and MLLABR is the average

labor intensity over the previous four quarters, where labor intensity is measured as one minus

property, plant, and equipment deflated by total assets.

Following Matsumoto [2002], we also expect MBE to be negatively correlated with

QLOSS, an indicator variable that takes value 1 if the firm reports loss for every quarter in the

previous four quarters and 0 otherwise; positively correlated with GROWEST, the I/B/E/S

median analyst estimate for long-term earnings growth measured at the end of the previous

year;20 positively correlated with POSUE, an indicator variable that equals to 1 if the earnings of

the quarter is greater than that of the same quarter in the previous year, and 0 otherwise. Finally,

we expect larger firms (measured as QLOGMV, log of market value at the end of the quarter) and

firms with lower uncertainty in forecasting environment (measured as AFE0, the deflated

absolute value of analyst forecast error at the beginning of the quarter) to be more likely to meet

or beat analyst consensus. We also include industry dummy variables (INDi) to control for

industry-specific factors that affect the industry’s overall likelihood of meeting or beating analyst

consensus. Because of the inclusion of the industry dummies, we omit the following industry-

specific variables from Matsumoto’s model: indicators for durable goods industry and high

litigation risk industry, and industry-specific measure for value relevance of earnings.

The results of the logistic regression estimation are presented in Panel C of Table 3. As

expected, the coefficient on DEDICATED is significantly positive at 1.139 with p-value less than

0.01, suggesting that even after controlling for determinants of the probability, dedicated guiders

19 Similar to Matsumoto [2002], when firms do not report quarterly, but report annual R&D information, we useone-fourth of the annual R&D for each of the four quarters during the corresponding fiscal year. In addition, we setR&D to zero if the variable is missing in both quarterly and annual Compustat databases.20 We use median, as opposed to mean, analyst long-term growth estimate because I/B/E/S recommends its use dueto the variance in methodologies for the calculation of long-term growth estimate among analysts. However, ourresults are qualitatively unchanged when we use mean analyst long-term growth estimate instead.

30

meet or beat analyst consensus significantly more frequently than occasional guiders. All of the

control variables have significant coefficients consistent with our expectations as described

above.

We argue that the greater propensity to “walk-down” market expectations and meet short-

term earnings targets is simply another observable aspect of the myopic focus of dedicated

guiders on achieving short-term goals. However, prior studies on expectations management have

documented that firms use earnings guidance to achieve beatable market expectations (e.g.,

Matsumoto [2002], Cotter et al. [2004]). Therefore it is unsurprising that dedicated guiders are

able to meet/beat expectations more often; it is a direct consequence of guiding the markets more

frequently. In order to distinguish our results from this alternative explanation, we re-estimate

Model (3) using quarters when firms do not provide any earnings guidance. Untabulated results

suggest dedicated guiders are more likely to meet/beat analysts’ forecasts even in quarters during

which they do not issue earnings guidance. These results show that the ability of dedicated

guiders to meet/beat earnings expectations is a cross-sectional phenomenon that likely reflects

the underlying myopic propensity of such firms rather than a mechanical consequence of

walking down expectations.

5.3. MEETING/BEATING ANALYST FORECASTS AND R&D EXPENDITURE

We next provide a direct and simple test of our myopia hypothesis by integrating our

prior analyses with respect to R&D spending and meeting/beating earnings expectations.

Specifically, we examine whether dedicated guiders are more likely to meet/beat earnings

expectations because they reduce R&D spending. To implement this, we separately compare the

“inherent propensity” of dedicated or occasional guiders to meet or beat earnings expectations

31

assuming the firm had maintained its previous R&D levels with the actual incidence of

meeting/beating expectations. Specifically, we construct a variable ADJMBE that indicates

whether a firm would have met or beaten consensus analysts’ earnings forecasts (measured prior

to earnings announcements) had it maintained the same level of R&D expenditure as in the same

quarter of the previous year.21 We then interact ADJMBE with the realized MBE, and report the

resulting contingency table.

Table 4 reports results of this analysis. Panel A reports the contingency table. First we

examine the columns where ADJMBE=1, i.e., where firms would have beaten earnings

expectations had they maintained R&D spending at previous levels. We observe that dedicated

guiders have a somewhat higher propensity to meet/beat expectation (i.e., MBE =1) than

occasional guiders (95% versus 88%), although the difference does not appear to be large.

Turning to columns where ADJMBE=0 (i.e., where firms would have missed earnings

expectations had they maintained R&D spending at previous levels) we see that a 64% of

dedicated guiders end up meeting or beating market expectations compared to 44% of occasional

firms. These results suggest that while both types of firms appear to reduce R&D to avoid

missing earnings expectations, dedicated guiders are far more aggressive in doing so.

Panel B presents the results in a format that better articulates the differences between

dedicated and occasional guiders. The first three columns examine the relative proportions of the

two types of firms meeting/beating expectations given prior R&D spending levels. Dedicated

guiders are 2.60 times as likely to meet/beat expectations than miss expectations based on prior

R&D levels versus 1.71 times for occasional guiders. That is, if maintaining prior R&D levels,

dedicated guiders are 1.52 (2.60/1.71) times more likely than occasional guiders to meet/beat

21 We note that the results are qualitatively similar if we construct ADJMBE based on R&D levels in the previousquarter.

32

than miss earnings expectations. This difference could arise because of inherent differences in

performance or better expectations management on the part of dedicated guiders. The last three

columns examine the relative proportions of the two types of firms meeting/beating expectations

given current R&D spending levels. Dedicated guiders are 6.24 times as likely to meet/beat

expectations than miss expectations using current R&D levels versus 2.51 times for occasional

guiders. That is, based on current R&D levels, dedicated guiders are 2.49 (2.60/1.71) times more

likely than occasional guiders to meet/beat than miss earnings expectations. Clearly, the

probability of meeting/beating expectation increases significantly for both dedicated and

occasional guiders when we use current R&D levels, indicating that both types of firms engage

in earnings management using R&D spending. However, the extent of cutting R&D for the

purpose of meeting/beating market expectations among dedicated firms is significantly more

aggressive.

In sum, we show that dedicated guiders more aggressively reduce R&D spending levels

to meet/beat earnings expectation than occasional guiders. This evidence suggests that dedicated

firms are more likely to sacrifice long-term benefits (i.e., R&D investments) for the sake of

meeting short-term goals (i.e., meeting/beating earnings expectations). This evidence is

consistent with the myopia hypothesis.

6. Guidance Frequency and Long-Term Performance

Our final hypothesis H3A examines the implications of quarterly guidance frequency on

firms’ long-term operating performance. Our measure of long-term operating performance is

change in return on assets (CROA) from 2003 to 2005. Since R&D expenditures decrease the

reported net income, in measuring CROA, we adjust for the effects of R&D expenditures by

33

adding back the after-tax R&D expenditure to net income before extraordinary items and then

deflate it by lagged total assets. We first report univariate analysis in Panel A of Table 5. The

mean CROA is 4.5% for dedicated guiders, in comparison to 4.1% for occasional guiders. While

this is inconsistent with our expectations, the difference is not statistically significant (p = 0.58).

Both groups of firms have median CROA of 3.1%; again the difference is not discernibly

different from zero (p = 0.84).

It is important to note that univariate analysis does not take into account other factors that

may potentially affect a firm’s long-term earnings growth. This is especially problematic when

these factors are correlated with guidance frequency. To properly explore the relationship

between guidance frequency and the long-term changes in return on assets, we need to

adequately address this correlated-omitted variable problem. Accordingly, we estimate the

following OLS regression that controls for other potential determinants of changes in ROA (firm

subscripts omitted):

CROA=0+iINDi+1DEDICATED+ 2BM2003+3LOGMV2003+

4GROWEST2003+5ROA2003+6NMBE + (4)

CROA and DEDICATED are as defined previously. We again include industry dummies (INDi)

to control for industry-specific factors that affect the long-term earnings growth. Book-to-market

ratio (BM2003), firm size (LOGMV2003), and median analyst forecast for long-term growth

(GROWEST2003) are measured at the end of fiscal year 2003 (indicated with subscripts 2003) in

order to control for economic conditions and market expectations for future growth at the

beginning of the period for which we measure the earnings growth. We expect firms with lower

BM2003, smaller LOGMV2003, or higher GROWEST2003 to have higher CROA over the subsequent

periods. We also include the level of ROA for 2003 (ROA2003) to control for the fact that firms

with higher ROA already have less growth opportunities and hence are expected to have lower

34

growth in operating performance. Finally, NMBE is the number of quarters, out of the twelve

quarters in 2001-2003, during which the firm meets or exceeds market expectations of earnings.

We include this variable because Bartov et al. [2002] show that firms that meet or beat analyst

forecasts have better operating performance in the future.

Table 5 Panel B reports the results of the OLS regression for Model (4). As predicted,

DEDICATED has a negative coefficient of -0.011, which is significant at the 0.02 level. This

magnitude also appears economically significant. The effect of firm size is insignificant. All

other variables have significant coefficients with signs consistent with our predictions.

Specifically, book-to-market ratio and return-on-assets in 2003 are significantly negatively

correlated with the change in ROA while analyst long-term growth forecasts and firms’

propensity in meeting or beating market expectations are significantly positively correlated with

the change in ROA.

To summarize, consistent with the myopia hypothesis, our results suggest that future

ROA growth is significantly lower for dedicated guiders than for occasional guiders, after

controlling for other determinants of changes in ROA. Thus, the empirical evidence in Sections

4-6 collectively supports our hypothesis that frequent quarterly earnings guidance is associated

with managerial myopic behavior, i.e., sacrificing long-term performance for the purpose of

meeting short-term objectives.

7. Conclusion

Recent discontinuance of quarterly earnings guidance by many firms has ignited a debate

about the costs and benefits of providing such guidance (e.g., McKay and Brown 2002). At issue

are the implications of earnings guidance for myopic managerial behavior. For example,

35

managers of companies discontinuing earnings guidance have remarked that earnings guidance

precipitates myopic decisions by forcing managers to emphasize meeting short-term earnings

targets rather than long-term value creation.

We provide empirical evidence that firms that frequently issue quarterly earnings

guidance exhibit relatively more myopic behavior than those that provide less frequent guidance.

Specifically, we document that dedicated guiders invest less in R&D, meet or beat analyst

consensus forecasts more frequently and have significantly lower long-term ROA growth than

occasional guiders. In particular, our result that dedicated guiders invest less in R&D is robust to

controlling for endogeneity between R&D investments and guidance frequency. Overall, our

evidence lends support to the argument that frequent quarterly earnings guidance fosters myopic

managerial behavior.

36

REFERENCES

AJINKYA, B., AND M. Gift. “Corporate Managers’ Earnings Forecasts and SymmetricalAdjustments of Market Expectations.” Journal of Accounting Research 22 (1984): 425-444.

AMEMIYA, T. “The Estimation of a Simultaneous Equation Generalized Probit Model.”Econometrica 46 (1978): 1193-1205.

BABER, W. R., P. M. FAIRFIELD, AND J. A. HAGGARD. “The Effect of Concern aboutReported Income on Discretionary Spending Decisions - The Case of Research-and-Development.” The Accounting Review 66 (1991): 818-829.

BARTOV, E., D. GIVOLY, AND C. HAYN. “The Rewards to Meeting or Beating EarningsExpectations.” Journal of Accounting & Economics 33 (2002): 173-204.

BEATTY, A., S. CHAMBERLAIN, AND J. MAGLIOLO. “Managing Financial Reports ofCommercial Banks: The Influence of Taxes, Regulatory Capital and Earnings.” Journal ofAccounting Research 33 (1995): 157-178.

BERGER, P. “Explicit and Implicit Tax Effects of the R&D Tax Credit.” Journal of AccountingResearch 31 (1993): 131-171.

BHOJRAJ, S., AND R. LIBBY. “Capital Market Pressure, Disclosure Frequency-InducedEarnings/Cash Flow Conflicts, and Managerial Myopia.” The Accounting Review 80 (2005): 1-20.

BHOJRAJ, S., P. HRIBAR, AND M. PICOONI. “Making Sense of Cents: An Examination ofFirms that Marginally Miss or Beat Analyst Forecasts.” Working Paper, Cornell University,2005.

BROWN, L., AND M. CAYLOR. “A Temporal Analysis of Thresholds: Propensities andValuation Consequences.” The Accounting Review 80 (2005): 423-440.

BUSHEE, B. “The Influence of Institutional Investors on Myopic R&D Investment Behavior.”The Accounting Review 73 (1998): 305-333.

CHEN, S., D. MATSUMOTO, AND S. RAJGOPAL. “Is Silence Golden? An EmpiricalAnalysis of Firms that Stop Giving Quarterly Earnings Guidance in the Post Regulation-FDPeriod.” Working paper, University of Washington, 2005.

CHENG, Q., AND T. D. WARFIELD. “Equity Incentives and Earnings Management.” TheAccounting Review 80 (2005) 441-476.

COTTER, J., A. I. TUNA, AND P.D. WYSOKI. “Expectations Management And BeatableTargets: How Do Analysts React to Explicit Earnings Guidance?” Working paper, Universityof South Queenthland, 2004.

DECHOW, P. M., AND R. G. SLOAN. “Executive Incentives and the Horizon Problem: AnEmpirical Investigation.” Journal of Accounting and Economics, 14 (1991): 51-89.

DEFOND, M. L., K. RAGHUNANDAN, AND K.R. SUBRAMANYAM. “Do Non=AuditService Fees Impair Auditor Independence? Evidence from Going Concern Audit Opinions.”Journal of Accounting Research 40 (2002): 1247-1274.

37

ELLIS, D. Capital: The Story of Long-Term Investment Excellence. Hoboken, N.J.: John Wiley& Sons, 2004.

FAMA, E. F., AND K. R. FRENCH. “Industry Costs of Equity.” Journal of FinancialEconomics 43 (1997): 153-193.

FAZZARI, S., G. HUBBARD AND B. PETERSEN. “Financing Constraints and CorporateInvestment,” Brookings Papers on Economic Activity (1988): 141–95.

FUDENBERG, D., AND J. TIROLE. “A Theory of Income and Dividend Smoothing Based onIncumbency Rents”, The Journal of Political Economy 103 (1995): 75-93.

FULLER J., AND M. JENSEN. “Just Say No to Wall Street.” Journal of Applied CorporateFinance 14 (2002): 41-46.

GOMPERS, P. A., AND A. METRICK. “Institutional Investors and Equity Prices.” QuarterlyJournal of Economics CXIV (2001): 229-260.

GRAHAM, J. R., C. R. HARVEY, AND S. RAJGOPAL. “The Economic Implications ofCorporate Financial Reporting.” Journal of Accounting and Economics 40 (2005): 3-73.

HALL. B., AND K. MURPHY. “The Trouble With Stock Options.” Journal of EconomicPerspectives 17 (3): 49-70.

HEALY, P., AND K. PALEPU. “Information Asymmetry, Corporate Disclosure, and the CapitalMarkets: A Review of the Empirical Disclosure Literature.” Journal of Accounting &Economics 31 (2001): 405-440.

HEFLIN, F., K. R. SUBRAMANYAM, AND Y. ZHANG. “Regulation FD and the FinancialInformation Environment: Early Evidence.” The Accounting Review 78 (2003): 1-37.

HOUSTON, J. F., B. LEV, AND J. TUCKER. “To Guide or Not to Guide? Causes andConsequences of Stopping and Subsequently Resuming Earnings Guidance.” Working paper,University of Florida, 2006.

JACOBS, M. Short-Term America: The Causes and Cures of Our Business Myopia. HarvardBusiness Press: Boston, MA, 1991.

JENSEN, M., AND W. MECKLING. “Theory of the Firm: Managerial Behavior, Agency Costsand Ownership Structure.” Journal of Financial Economics 4 (1976): 305-360.

KAPLAN, S. N. AND L. ZINGALES. Do investment-cash flow sensitivities provide usefulmeasures of financing constraints? Quarterly Journal of Economics (1997) 113:169–215.

KASZNIK, R. “On the Association between Voluntary Disclosure and Earnings Management.”Journal of Accounting Research, 37 (1999): 57-81.

KASZNIK, R., AND M. MCNICHOLS. “Does Meeting Earnings Expectations Matter?Evidence from Analyst Forecast Revisions and Share Prices.” Journal of Accounting Research40 (2002): 727-759.

LAFOND, R., AND R. L. WATTS. “The Information Role of Conservative FinancialStatements.” Working paper, MIT, 2006.

LANG, M., AND R. LUNDHOLM. “Cross-Sectional Determinants of Analyst Ratings ofCorporate Disclosures.” Journal of Accounting Research 31 (1993): 246-71.

38

LARCKER, D. F., and T. O. RUSTICUS. “On the Use of Instrumental Variables in AccountingResearch.” Working paper, Wharton School, University of Pennsylvania (2005).

LEVITT. “Renewing the Covenant with Investors.” May 10, 2000. New York.http://www.sec.gov/news/speech/spch370.htm

LEV, B., AND S. PENMAN. “Voluntary Forecast Disclosures, Nondisclosure, and StockPrices.” Journal of Accounting Research 28 (1990): 49-76.

MATSUNAGA, S.R., AND C. W. PARK. “The Effect of Missing a Quarterly EarningsBenchmark on the CEO’s Annual Bonus.” The Accounting Review 76 (2001): 313-332.

MATSUMOTO, D. “Management’s Incentives to Avoid Negative Earnings Surprises.” TheAccounting Review 77 (2002): 483-514.

MYERS, S. “The Capital Structure Puzzle.” Journal of Finance 39 (1984): 575-592.

MCKAY, B., AND K. BROWN. “Leading the News: Coke to Abandon Profit Forecasts, ProvideData on Long-Term Goals.” Wall Street Journal December 16, 2002.

NELSON, F., AND L. OLSEN. “Specification and Estimation of Simultaneous Equations Modelwith Limited Dependent Variables.” International Economic Review 19 (1978): 695-710.

PORTER, M. E. “Capital Choices: The Causes and Cures of Business Myopia.” Research Reportto the U.S. Government’s Council on Competitiveness, Washington D.C., 1992.

PRINCE, C.J. “Just Say No?” Entrepreneur (August 2005): 54-55.

RAPPAPORT, A. “The Economics of Short-Term Performance Obsession.” Financial AnalystsJournal 61 (2005): 65-79.

ROYCHOWDHURY, S. “Earnings Management through Real Activities Manipulation” Journalof Accounting and Economics 42 (2006): 335-370.

SKINNER, D. J. “Why Firms Voluntarily Disclose Bad News.” Journal of Accounting Research32 (1994): 38-60.

SKINNER, D. J. “Earnings Disclosures and Stockholder Lawsuits.” Journal of Accounting andEconomics 23 (1997): 249-282.

SKINNER, D. J., AND R. SLOAN. “Earnings Surprises, Growth Expectations, and StockReturns or Don’t Let an Earnings Torpedo Sink Your Portfolio.” Review of Accounting Studies7 (2002): 289-312.

SLOAN, R.G. “Do Stock Prices Fully Reflect Information in Cash Flows and Accruals AboutFuture Earnings?” The Accounting Review (1996): 289-315.

STEIN, J. C. “Takeover Threats and Managerial Myopia.” Journal of Political Economy 96(1988): 61-79.

STEIN, J. C. “Efficient Capital Markets, Inefficient Firms: A Model of Myopic CorporateBehavior.” The Quarterly Journal of Economics 104 (1989): 655-669.

TEOH, S. H. I. WELCH, AND T. J. WONG. “Earnings Management and the Underperformanceof Seasoned Equity Offerings.” Journal of Financial Economics 50 (1998): 63-99.

39

Figure 1: The Endogenous Nature of Earnings Guidance

Poor FinancialPerformance

Litigation Risk

TransientInvestors

Managerial MyopiaSub-optimal Underinvestment in

Long Term Value Creation

FinancialConstraints

Market Focus onShort Term

Earnings

Pressure to meetShort-Term

Earnings Targets

Short-Term ManagerialMotivations

(Manager’s horizon, jobconcerns, incentive

compensation, need toraise capital)

Managerial Focuson Short-Term

Earnings

Dedicated QuarterlyEarnings Guidance

40

Table 1: Description of Quarterly Earnings Guidance Frequency

Industry # firms Mean Median Std. Dev. Q1 Q3 # firmsFREQ=0

# firmsFREQ=12

Overall 1965 4.02 3 3.76 1 7 480 72Business Equipment 557 4.64 4 3.56 2 7 75 21Chemicals 51 5.00 5 3.97 1 8 6 5Consumer Durables 63 4.59 4 3.71 1 8 10 4Consumer Non-Durables 107 4.70 4 3.96 1 8 20 6Energy 98 1.98 0 3.14 0 3 53 2Healthcare 249 2.12 1 3.00 0 3 117 3Manufacturing 257 3.95 3 3.54 1 7 54 6Telecommunications 67 2.06 0 3.20 0 3 34 2Wholesale and Retails 216 5.93 7 4.12 2 9 32 18Other 300 3.72 2 3.67 0 7 79 5

Table 1 reports descriptive statistics of quarterly earnings guidance frequency from 2001 to 2003 for a sample of 1,965 firms.Quarterly earnings guidance frequency is defined as the number of quarters for which a firm issues at least one earnings guidance forthe quarter. The industry classification is according to the 12 categories in Fama and French (1997), with the utilities industry and thefinancial service industry excluded.

41

Table 2: Tests for Research and Development Expenditures

Panel A: Univariate Analyses

Means Mediansdedicated occasional dedicated occasional

RDX 0.047 0.084 0.012 0.013p-value 0.00 p-value 0.00

ADJRDX -0.018 0.016 -0.010 -0.004p-value 0.00 p-value 0.00

Panel B: OLS Regression Analyses

Dep. Var. = RDX; # Obs. = 2778Independent Variables Predicted Sign Coefficient p-value

IND * * *Intercept 0.022 0.00DEDICATED - -0.004 0.02LAGRDX + 0.093 0.00LAGDEDI ? 0.001 0.22BM - -0.006 0.00LOGMV - 0.000 0.90EP - 0.008 0.00AGE - -0.003 0.04ADJROA + 0.207 0.00NADJROA ? -0.089 0.00LOSS - 0.036 0.00CAPX - -0.043 0.00ADJFCF + 0.002 0.85DILUTION - -0.051 0.03EQOFFER - 0.050 0.00%TINST - -0.025 0.00NUMANA + 0.000 0.91VALRE - -0.002 0.00Adj-Rsq 68.47%

42

Panel C: Two-Stage System of Equations Analyses

DEP

INDEP

Two-Equation System; #Obs.=2655RDX MODEL DEDICATED MODEL

Sign Coeff. p-value Sign Coeff. p-valueIND * * * * * *

Intercept 0.053 0.00 1.728 0.00

PREDDEDI - -0.048 0.00

PREDRDX ? -2.504 0.30

LAGRDX + 0.090 0.00

LAGDEDI ? 0.013 0.00 + 1.498 0.00

BM - -0.009 0.00

LOGMV - -0.001 0.25 + -0.334 0.00

EP - 0.012 0.00 - 0.341 0.09

AGE - -0.004 0.00

ADJROA + 0.197 0.00

NADJROA ? -0.107 0.00

LOSS - 0.031 0.00 ? -0.537 0.00

CAPX - -0.039 0.01

ADJFCF + 0.021 0.11

DILUTION - -0.030 0.22 + 2.285 0.11

EQOFFER - 0.044 0.00 + -0.241 0.30

%TINST - -0.001 0.91 + 4.021 0.00

NUMANA + 0.001 0.00 + 0.211 0.00

VALRE - -0.003 0.03 + 0.409 0.00

MAFE0 + -5.722 0.14

MSTD0 + -1.007 0.00

PIN + -0.549 0.38

EXPDAM ? 0.000 0.16Adj-Rsq 68.76% 49.73%

Table 2 Panel A reports univariate comparisons for R&D expenditures (RDX) and industry-adjusted (ADJRDX) expenditures between dedicated guiders and occasional guiders. Dedicatedguiders (occasional guiders) refer to firms that are in the top (bottom) third of their respectiveindustry’s distribution of guidance frequency. Numbers in parentheses are two-sided p-values.For means, the p-values are based on t-tests; for medians, the p-values are based on Wilcoxon

43

tests. Panel B reports the coefficients and two-sided p-values of the ordinary least squareregressions of R&D expenditures (i.e., Model (1)). Panel C reports the coefficients and two-sidedp-values of the second-stage estimation of the two-stage systems of equations of R&Dexpenditures and guidance frequency determinants (i.e., Models (1) and (2) together).The variables are defined as follows:RDX = R&D expenses (Compustat data#46) for the year scaled by total assets

(Compustat data#6) at the beginning of the fiscal year,IND = industry dummy variables,DEDICATED = one (zero) if the firm’s quarterly earnings guidance frequency is in the top

(bottom) third of their respective industry’s distribution of guidancefrequency based on the observation from year 2001 to 2003,

PREDDEDI = the predicted value of dedication, which is the fitted value (i.e., a continuousmeasure) from the first-stage logistic regression of DEDICATED on allexogenous variables from Model (1) and Model (2) together,

PREDRDX = the predicted value of R&D expenses, which is the fitted value from thefirst-stage ordinary least square regression of RDX on all exogenousvariables from Model (1) and Model (2) together,

LAGRDX = R&D expenses (Compustat data#46) at the end of year 2000 scaled by totalassets (Compustat data#6) at the beginning of year 2000,

LAGDEDI = the rank of the sample firms’ quarterly earnings guidance frequency (intothree groups) within their respective industry’s distribution based on datafrom 1998 to 2000,

BM = ratio of book value of equity (Compustat data#60) to market value of equity(Compustat data#25 x data#199) at the end of the year 2000,

LOGMV = log of market value at the end of the year 2000,EP = earnings per share (Compustat data#58) divided by price per share

(Compustat data#199) at the end of the year,AGE = log of firm age,ADJROA = R&D-adjusted return on assets measured as net income before extraordinary

items (Compustat data#18) plus 0.65 times R&D expenses (Compustatdata#46) divided by total assets (Compustat data#6) at the beginning of theyear,

NADJROA = ADJROA if ADJROA is negative and 0 otherwise,LOSS = 1 if net income before extraordinary items (Compustat data#18) is negative

and 0 otherwise,CAPX = capital expenditure (Compustat data#30) for fiscal year scaled by total

assets (Compustat data#6) at the beginning of the fiscal year,ADJFCF = R&D adjusted free cash flow measured using income (Compustat

data#18) minus the change of current assets (Compustat data#4) plus thechange of current liabilities (Compustat data#5) plus the change of cash andshort-term investments (Compustat data#1) minus the change of short-termdebt (Compustat data#34) plus depreciation (Compustat data#14) minuscapital expenditure (Compustat data#30) plus 0.65 times R&D expenses(Compustat data#46) scaled by total assets (Compustat data#6) at thebeginning of the fiscal year,

DILUTION = the difference between basic earnings per share (Compustat data#58) and

44

diluted earnings per share (Compustat data#57) deflated by the dilutedearnings per share (Compustat data#57) at the end of the year,

EQOFFER = total net value of new equity issuance (Compustat data#108) during 2001 to2003 scaled by total assets at the end of the year 2000,

%TINST = percentage of transient institutional holding at the end of the year fromCDA/Spectrum Institutional (13f) Holdings database according to Bushee(1998),

NUMANA = number of analysts following at the end of the calendar year from First CallSummary database,

VALRE = value relevance proxy, measured as negative one times the squared residualfrom the following regression:

0 1 2 3 4*RET NIBE LOSS NIBE LOSS NIBE .The regression is estimated by three, two, or one-digit SIC code conditionalon having at least 10 firms in each group. RET = market adjusted returnover the fiscal year; NIBE = net income before extraordinary items(Compustat annual data#18) scaled by the beginning market value of equity(Compustat annual data#25*Compustat annual data#199); LOSS =1 ifNIBE is negative, zero otherwise; NIBE = the change in net income beforeextra-ordinary items scaled by beginning market value of equity. Since wemultiply by -1 larger values of VALREL indicates greater value relevance,

MAFE0 = average absolute analyst forecast error measured as the absolute value of thedifference between actual earnings per share and the mean forecast at thebeginning of the quarter scaled by price at the quarter end over the currentyear,

MSTD0 = average standard deviation of analyst forecasts measured as the standarddeviation of the beginning of the quarter forecasts scaled by price at thequarter end over the current year,

PIN = a measure of probability of information-related trading estimated using theEasley, Hvidkjaer and O'hara (2002) trading model for the current year,

EXPDAM = expected loss damage following Skinner (1997) to measure potentiallitigation costs calculated as –MVE*MKTADJRET*(1-(1-VOL)**N));MVE=market value at the beginning of the year; MKTADJRET=marketadjusted return over the year; VOL=average percentage of the firm’s sharestraded over the year; N=day period between the end of the year and thebeginning of the year.

45

Table 3: Tests for Quarterly Performance: Meeting/Beating Analyst Forecast

Panel A: Univariate Analysis

Means Mediansdedicated occasional dedicated occasional

MBE0 0.53 0.51 1.00 1.00p-value 0.02 p-value <0.02

MBE 0.85 0.71 1.00 1.00p-value 0.00 p-value 0.00

Panel B: Quarterly Earnings Guidance and Meeting/Beating Analyst Forecasts

MBE0=0 MBE0=1MBE=0 MBE=1 MBE=0 MBE=1

Dedicated 1982 (30%) 4688 (70%) 175 (2%) 7229(98%)

Occasional 3516 (53%) 3159 (47%) 410 (6%) 6652(94%)

p-value 0.00 p-value 0.00

Panel C: Logistic Regression Analyses of Meeting/Beating Analyst ForecastsDep. Var. = MBE; # Obs. = 17,015Independent Variables Predicted Sign Coefficient p-valueIND * * *DEDICATED + 1.139 0.00ML%INST + 0.572 0.00MLRDX + 5.399 0.00MLLABR + 0.541 0.00QLOSS - -0.534 0.00GROWEST + 0.009 0.00POSUE + 2.566 0.00QLOGMV + 0.136 0.00AFE0 - -37.061 0.00Adj-rsq 41.35%

Table 3 Panel A reports the univariate comparisons for the quarterly meeting/beating outcomesat the beginning of the quarter (MBE0) and at the end of the quarter (MBE) between dedicatedguiders and occasional guiders. Dedicated guiders (occasional guiders) refer to firms that are inthe top (bottom) third of their respective industry’s distribution of guidance frequency. The p-values are two-sided. Tests for means are based on t test and for medians are based on Wilcoxontest. Panel B reports the interaction between the frequency of meeting/beating analyst forecasts atthe beginning of the quarter and that at the earnings announcement for dedicated guiders andoccasional guiders separately. All variables are based on firm-quarter-specific observations.Panel C reports the coefficients and two-sided p-values of the logit regressions of the probabilityof meeting or beating analyst forecasts. The variables are defined as follows:

46

MBE0 = one if actual earnings per share is no less than the analyst forecast meanoutstanding at the beginning of the quarter deflated by beginning-of-quarterstock price, and zero otherwise,

MBE = one if quarterly actual earnings per share is greater than or equal to the lastanalyst forecast mean for the quarter, and zero otherwise,

DEDICATED = one (zero) if the firm’s quarterly earnings guidance frequency is in the top(bottom) third of their respective industry’s distribution of guidancefrequency based on the observation from year 2001 to 2003,

IND = industry dummy variables,ML%INST = percentage of shares held by institutional investors averaged over the

previous four quarters,MLRDX = research and development expenses (Quarterly Compustat data#4) over total

assets (Quarterly Compustat data#44) averaged over the previous fourquarters,

MLLABR = one minus property, plant, and equipment (Quarterly Compustat data#42)divided by total assets (Quarterly

Compustat data#44) averaged over the previous four quarters,QLOSS = one if quarterly net income (Quarterly Compustat data#8) in each of the

previous four quarters is negative,and zero otherwise,

GROWEST = median of analysts’ long-term growth forecasts at the end of the year fromIBES Summary file,

POSUE = one if earnings (Quarterly Compustat data#8) in the quarter is greater thanearnings of the same quarter in

the prior year, and zero otherwise,QLOGMV = log of market value of equity (Quarterly Compustat data#61 x data#14) at

the end of the quarter,AFE0 = absolute value of quarterly actual earnings per share minus the last analyst

forecast mean outstanding at thebeginning of the quarter deflated by beginning-of-quarter stock price.

47

Table 4: Linking R&D Expenditure and Meeting/Beating Analysts’ Consensus Forecasts

Panel A: Interaction between DEDICATED, ADJMBE and MBE

ADJMBE=0 ADJMBE=1MBE=0 MBE=1 MBE=0 MBE=1

Dedicated 406 (36%) 715 (64%) 152 (5%) 2767 (95%)Occasional 865 (56%) 678 (44%) 328 (12%) 2314 (88%)

p-value 0.00 p-value 0.00

Panel B: Relative Frequency between Meeting/Beating and Missing Market Expectations

ADJMBE MBE=0 (a) =1 (b) (b)/(a) =0 (a) =1 (b) (b)/(a)

Dedicated 1121 (28%) 2919 (72%) 2.60 558 (14%) 3482 (86%) 6.24Occasional 1543 (37%) 2642 (63%) 1.71 1193 (29%) 2992 (71%) 2.51

Panel A and Panel B of Table 4 present different contingency tables among DEDICATED, MBE, and ADJMBE based on quarterlyobservations during 2001 and 2003. Dedicated guiders (occasional guiders) refer to firms that are in the top (bottom) third of theirrespective industry’s distribution of guidance frequency. MBE is an indicator variable that equals to 1 if the firm meets/beats marketexpectations at the time of earnings announcement, and 0 otherwise. ADJMBE is an indicator variable that equals to 1 if the firmmeets/beats market expectations at the time of earnings announcements, had the firm maintained its R&D spending level in the samequarter of prior year, and 0 otherwise.

48

Table 5: Tests for Long-Term Performance: Increase in ROA from 2000 to 2003

Panel A: Univariate Analyses

Means Mediansdedicated occasional dedicated occasional

CROA 0.045 0.041 0.031 0.031p-value 0.58 p-value 0.84

Panel B: Regression Analyses

Dep. Var. = CROA; # Obs. = 851Independent Variables Predicted Sign Coefficient p-valueIND * * *DEDICATED - -0.011 0.02BM2003 - -0.029 0.00LOGMV2003 - 0.001 0.47GROWEST2003 + 0.001 0.00ROA2003 - -0.245 0.00NMBE + 0.004 0.00Adj-Rsq 16.17%

Table 5 Panel A reports univariate analyses for change in ROA between dedicated guiders and occasional guiders from 2003 to 2005.Dedicated guiders (occasional guiders) refer to firms that are in the top (bottom) third of their respective industry’s distribution ofguidance frequency. Numbers in parentheses are two-sided p-values. For means, the p-values are from t test; for medians, the p-valuesare from Wilcoxon test. Panel B reports the coefficients and two-sided p-values of ordinary least square regressions of change inROA. All variables are based on one observation per firm. The variables are defined as follows:

DEDICATED = one (zero) if the firm’s quarterly earnings guidance frequency is in the top (bottom) third of their respectiveindustry’s distribution of guidance frequency based on the observation from year 2001 to 2003,

IND = industry dummy variables,CROA = change in net income before extraordinary items (Compustat data#18) from fiscal year 2003 to 2005 deflated by

total assets (data#6) at the end of fiscal year 2003,

49

BM2003 = ratio of book value of equity (Compustat data#60) to market value of equity (Compustat data#25 x data#199) atthe end of fiscal year 2003,

MV2003 = market value (Compustat data#25 x data#199) at the end of fiscal year 2003,LOGMV2003 = log of market value at the end of fiscal year 2003,GROWEST2003 = median of analysts’ long-term growth forecasts within three months prior to fiscal year end of 2003 from

IBES Summary file,ROA2003 = return on assets in 2003 (Compustat data#18 / data #6),NMBE = number of quarters meeting or beating quarterly analyst forecasts during 2001-2003.